Advanced search

Message boards : Number Crunching : Monster MNIST units

Previous · 1 · 2 · 3 · 4 · Next
Author Message
Profile Steve Dodd
Send message
Joined: 24 Jul 13
Posts: 10
Combined Credit: 17,373,918
DNA@Home: 22,044
SubsetSum@Home: 1,017,427
Wildlife@Home: 16,334,447
Wildlife@Home Watched: 994s
Wildlife@Home Events: 0
Climate Tweets: 44
Images Observed: 0

        
Message 6879 - Posted: 7 Apr 2017, 17:40:23 UTC

Oh well, might as well add my 2 cents worth - http://csgrid.org/csg/result.php?resultid=2142073
____________

Travis Desell
Volunteer moderator
Project administrator
Project developer
Project scientist
Send message
Joined: 16 Jan 12
Posts: 1813
Combined Credit: 23,514,257
DNA@Home: 293,563
SubsetSum@Home: 349,212
Wildlife@Home: 22,871,482
Wildlife@Home Watched: 212,926s
Wildlife@Home Events: 51
Climate Tweets: 21
Images Observed: 774

              
Message 6880 - Posted: 7 Apr 2017, 19:53:28 UTC - in response to Message 6878.

Hello Travis,

I have now a Problem with a Monster-WU too, I crunch now 4,5 Days on this WU and I wait some Hours for the okay of my Wingman. Now I see I become only 1092 Credits, ist very crazy Travis.
If you want check my WU, so I can give you this Informations from me. :

2136329 979728 32305 2 Apr 2017, 6:12:10 UTC 6 Apr 2017, 9:41:13 UTC Fertig und Bestätigt 358,028.01 286,172.30 1,091.76 EXACT MNIST Convolutional Neural Network Trainer v0.20

Many Greetz SEARCHER


This was most likely sent out before my credit update. I'm hoping the change should fix this.

Profile Conan
Avatar
Send message
Joined: 13 Apr 12
Posts: 151
Combined Credit: 47,672,899
DNA@Home: 399,792
SubsetSum@Home: 1,448,876
Wildlife@Home: 45,824,231
Wildlife@Home Watched: 70,910s
Wildlife@Home Events: 0
Climate Tweets: 412
Images Observed: 0

          
Message 6881 - Posted: 7 Apr 2017, 22:28:38 UTC - in response to Message 6877.

I've updated the my algorithm for calculating credit which I think should fix the credit inconsistencies. If you're still seeing this with newly generated workunits let me know.


I have one at 2 days 10 hours and 48.3% completed with 2 days 14 hours still to go.
As it was sent out on the 3rd I may not get the new credit but we will see how it travels. My wingman must be running a long time too as it is still running for them as well.

Conan

Has now reached 3 days 9 hours, 60.9% and 2 days 4 hours to go.

This one is going to be my longest, but as it was before Travis's credit update I probably wont get a lot for it.
Still helping the science, points or no points (would prefer some of those points though).

Conan
____________

Travis Desell
Volunteer moderator
Project administrator
Project developer
Project scientist
Send message
Joined: 16 Jan 12
Posts: 1813
Combined Credit: 23,514,257
DNA@Home: 293,563
SubsetSum@Home: 349,212
Wildlife@Home: 22,871,482
Wildlife@Home Watched: 212,926s
Wildlife@Home Events: 51
Climate Tweets: 21
Images Observed: 774

              
Message 6882 - Posted: 8 Apr 2017, 4:37:59 UTC - in response to Message 6881.

I've updated the my algorithm for calculating credit which I think should fix the credit inconsistencies. If you're still seeing this with newly generated workunits let me know.


I have one at 2 days 10 hours and 48.3% completed with 2 days 14 hours still to go.
As it was sent out on the 3rd I may not get the new credit but we will see how it travels. My wingman must be running a long time too as it is still running for them as well.

Conan

Has now reached 3 days 9 hours, 60.9% and 2 days 4 hours to go.

This one is going to be my longest, but as it was before Travis's credit update I probably wont get a lot for it.
Still helping the science, points or no points (would prefer some of those points though).

Conan


If it makes you feel any better, given the current progress of what we're doing:

http://csgrid.org/csg/exact/overview.php

We've hit 99.52% on the training data and with this genome we've hit 98.71% on the testing data (i.e., how well the neural network does on things it's never seen before). So we're starting to get close to the best trained neural networks for the MNIST dataset. Pretty cool considering the networks you guys and gals are training are evolved as opposed to architected by a human.

Profile SEARCHER
Avatar
Send message
Joined: 24 Feb 13
Posts: 29
Combined Credit: 1,450,269
DNA@Home: 121,784
SubsetSum@Home: 114,603
Wildlife@Home: 1,213,882
Wildlife@Home Watched: 0s
Wildlife@Home Events: 0
Climate Tweets: 3,592
Images Observed: 35,153

          
Message 6884 - Posted: 9 Apr 2017, 8:31:30 UTC

Good morning and hello Travis, thank you very much for your work.

Greetz SEARCHER
____________
Member of CHARITY TEAM

Profile scole of TSBT
Send message
Joined: 15 Nov 14
Posts: 3
Combined Credit: 22,242,267
DNA@Home: 885,086
SubsetSum@Home: 5,067,895
Wildlife@Home: 16,289,286
Wildlife@Home Watched: 0s
Wildlife@Home Events: 0
Climate Tweets: 0
Images Observed: 0

      
Message 6887 - Posted: 10 Apr 2017, 20:47:54 UTC

I've had a few run 30+ hours and this one has been running over 54 hours at 83.444% complete...http://csgrid.org/csg/workunit.php?wuid=995904

Profile Conan
Avatar
Send message
Joined: 13 Apr 12
Posts: 151
Combined Credit: 47,672,899
DNA@Home: 399,792
SubsetSum@Home: 1,448,876
Wildlife@Home: 45,824,231
Wildlife@Home Watched: 70,910s
Wildlife@Home Events: 0
Climate Tweets: 412
Images Observed: 0

          
Message 6888 - Posted: 11 Apr 2017, 6:15:35 UTC
Last modified: 11 Apr 2017, 6:16:38 UTC

OK finally finished this one 524,027.60 seconds WU 1000879, around 7 days or so.
Have just downloaded another couple of long ones estimated at over 8 days and 5 days.

Hope the credit is not too bad (it is before Travis's credit update) for the effort involved.

Conan
____________

Travis Desell
Volunteer moderator
Project administrator
Project developer
Project scientist
Send message
Joined: 16 Jan 12
Posts: 1813
Combined Credit: 23,514,257
DNA@Home: 293,563
SubsetSum@Home: 349,212
Wildlife@Home: 22,871,482
Wildlife@Home Watched: 212,926s
Wildlife@Home Events: 51
Climate Tweets: 21
Images Observed: 774

              
Message 6892 - Posted: 12 Apr 2017, 16:07:32 UTC - in response to Message 6888.

OK finally finished this one 524,027.60 seconds WU 1000879, around 7 days or so.
Have just downloaded another couple of long ones estimated at over 8 days and 5 days.

Hope the credit is not too bad (it is before Travis's credit update) for the effort involved.

Conan


If it is low let me know and I'll adjust the algorithm some more. Will keep working on this until we get it right. Sucks that the turnaround is a bit slow due to the size of the workunits however.

Zir Madmax
Send message
Joined: 17 Feb 17
Posts: 11
Combined Credit: 513,325
DNA@Home: 0
SubsetSum@Home: 0
Wildlife@Home: 513,325
Wildlife@Home Watched: 0s
Wildlife@Home Events: 0
Climate Tweets: 0
Images Observed: 0

  
Message 6895 - Posted: 12 Apr 2017, 19:04:11 UTC

(Reposted it on right place now maby)
Hi ... i have 1 monster .. at the begining i say 1 day and 8 houers Now after 1d.12h it say over 10d.2h days..
name: exact_genome_1491416078_5_8943_1
And ther is no posibility to make it finish till 04/17 It still ading some houer now and then.
This is not the first.. Befor i think it be a error on my pc so i make a clean install but now it start agin.
will u i delite it? or u can add some extra days for me? I need atleast 5 more days
Thomas/ Zir Madmax sweden

Profile Conan
Avatar
Send message
Joined: 13 Apr 12
Posts: 151
Combined Credit: 47,672,899
DNA@Home: 399,792
SubsetSum@Home: 1,448,876
Wildlife@Home: 45,824,231
Wildlife@Home Watched: 70,910s
Wildlife@Home Events: 0
Climate Tweets: 412
Images Observed: 0

          
Message 6896 - Posted: 12 Apr 2017, 22:02:44 UTC - in response to Message 6892.
Last modified: 12 Apr 2017, 22:39:09 UTC

OK finally finished this one 524,027.60 seconds WU 1000879, around 7 days or so.
Have just downloaded another couple of long ones estimated at over 8 days and 5 days.

Hope the credit is not too bad (it is before Travis's credit update) for the effort involved.

Conan


If it is low let me know and I'll adjust the algorithm some more. Will keep working on this until we get it right. Sucks that the turnaround is a bit slow due to the size of the workunits however.


It payed 1,496.97 points.

Not a great deal for WU length.
My wingman did the same work unit in 68,891.67 seconds, just a wee bit more than 3 times less than my result (it is 7 times the difference), even accounting for a faster processor the speed difference is wide.

My next really big one (as opposed to just a 4 day large one), has completed 2 days 21 hours at 28.476% and says it needs another 7 days 7 hours. So like the last big one is may also go past it's return date.

Thanks Travis

Conan
____________

Profile Conan
Avatar
Send message
Joined: 13 Apr 12
Posts: 151
Combined Credit: 47,672,899
DNA@Home: 399,792
SubsetSum@Home: 1,448,876
Wildlife@Home: 45,824,231
Wildlife@Home Watched: 70,910s
Wildlife@Home Events: 0
Climate Tweets: 412
Images Observed: 0

          
Message 6903 - Posted: 14 Apr 2017, 22:40:04 UTC
Last modified: 14 Apr 2017, 22:41:03 UTC

Had this one and both myself and my wingman had large run times but the credit seems a bit low for the run time this wu

Mine ran over 260,000 seconds and my wingman ran over 332,000 seconds for 1,911 credits.

Conan
____________

Travis Desell
Volunteer moderator
Project administrator
Project developer
Project scientist
Send message
Joined: 16 Jan 12
Posts: 1813
Combined Credit: 23,514,257
DNA@Home: 293,563
SubsetSum@Home: 349,212
Wildlife@Home: 22,871,482
Wildlife@Home Watched: 212,926s
Wildlife@Home Events: 51
Climate Tweets: 21
Images Observed: 774

              
Message 6905 - Posted: 15 Apr 2017, 1:56:04 UTC - in response to Message 6903.

Had this one and both myself and my wingman had large run times but the credit seems a bit low for the run time this wu

Mine ran over 260,000 seconds and my wingman ran over 332,000 seconds for 1,911 credits.

Conan


Thoughts on what might be more appropriate for that kind of runtime?

Travis Desell
Volunteer moderator
Project administrator
Project developer
Project scientist
Send message
Joined: 16 Jan 12
Posts: 1813
Combined Credit: 23,514,257
DNA@Home: 293,563
SubsetSum@Home: 349,212
Wildlife@Home: 22,871,482
Wildlife@Home Watched: 212,926s
Wildlife@Home Events: 51
Climate Tweets: 21
Images Observed: 774

              
Message 6906 - Posted: 15 Apr 2017, 1:56:30 UTC - in response to Message 6896.

OK finally finished this one 524,027.60 seconds WU 1000879, around 7 days or so.
Have just downloaded another couple of long ones estimated at over 8 days and 5 days.

Hope the credit is not too bad (it is before Travis's credit update) for the effort involved.

Conan


If it is low let me know and I'll adjust the algorithm some more. Will keep working on this until we get it right. Sucks that the turnaround is a bit slow due to the size of the workunits however.


It payed 1,496.97 points.

Not a great deal for WU length.
My wingman did the same work unit in 68,891.67 seconds, just a wee bit more than 3 times less than my result (it is 7 times the difference), even accounting for a faster processor the speed difference is wide.

My next really big one (as opposed to just a 4 day large one), has completed 2 days 21 hours at 28.476% and says it needs another 7 days 7 hours. So like the last big one is may also go past it's return date.

Thanks Travis

Conan


Do you think the 3x difference could be accounted for by a faster CPU? I don't suppose you have a link to that workunit?

Profile Conan
Avatar
Send message
Joined: 13 Apr 12
Posts: 151
Combined Credit: 47,672,899
DNA@Home: 399,792
SubsetSum@Home: 1,448,876
Wildlife@Home: 45,824,231
Wildlife@Home Watched: 70,910s
Wildlife@Home Events: 0
Climate Tweets: 412
Images Observed: 0

          
Message 6907 - Posted: 15 Apr 2017, 9:11:47 UTC - in response to Message 6905.
Last modified: 15 Apr 2017, 9:39:18 UTC

Had this one and both myself and my wingman had large run times but the credit seems a bit low for the run time this wu

Mine ran over 260,000 seconds and my wingman ran over 332,000 seconds for 1,911 credits.

Conan


Thoughts on what might be more appropriate for that kind of runtime?


Just going on what some of my other much shorter work units are getting then the awarded credit would have to be over 7,000 points (I have seen a number of work units around the 70,000 second mark get 5,000+ points, I have a few myself).
So 7,000 would be a lower amount for these much larger work units and then doing what I suggest below.

The multiplier on some of the short ones if it was applied against the larger ones would see you handing out 20,000 to 30,000 points a work unit which is getting a bit high don't you think?

I suppose you could just set a maximum limit you are willing to pay, say 10,000 points and no matter how long the work units then run they will get 10,000 points, which is still a good return.
You just would have to work out when that limit cuts in with a scaling of awarded points up to the limit then a set amount.

I have 4 large work units with very ordinary credit, ranging from 10 to 28 cr/h

1009561 98.82h / 2825.5cr = 28.59 cr/h
1008772 114.55h / 3218.04cr = 28.09 cr/h
1001360 92.46h / 1911.57cr = 20.67 cr/h
1000879 145.56h / 1496.97cr = 10.28 cr/h

When you get some of the short ones getting 300cr/h then the long ones should get a better deal.

Just some thoughts.
Thanks Travis

Conan
____________

Profile Conan
Avatar
Send message
Joined: 13 Apr 12
Posts: 151
Combined Credit: 47,672,899
DNA@Home: 399,792
SubsetSum@Home: 1,448,876
Wildlife@Home: 45,824,231
Wildlife@Home Watched: 70,910s
Wildlife@Home Events: 0
Climate Tweets: 412
Images Observed: 0

          
Message 6908 - Posted: 15 Apr 2017, 9:18:06 UTC - in response to Message 6906.
Last modified: 15 Apr 2017, 9:40:34 UTC

OK finally finished this one 524,027.60 seconds WU 1000879, around 7 days or so.
Have just downloaded another couple of long ones estimated at over 8 days and 5 days.

Hope the credit is not too bad (it is before Travis's credit update) for the effort involved.

Conan


If it is low let me know and I'll adjust the algorithm some more. Will keep working on this until we get it right. Sucks that the turnaround is a bit slow due to the size of the workunits however.


It payed 1,496.97 points.

Not a great deal for WU length.
My wingman did the same work unit in 68,891.67 seconds, just a wee bit more than 3 times less than my result (it is 7 times the difference), even accounting for a faster processor the speed difference is wide.

My next really big one (as opposed to just a 4 day large one), has completed 2 days 21 hours at 28.476% and says it needs another 7 days 7 hours. So like the last big one is may also go past it's return date.

Thanks Travis

Conan


Do you think the 3x difference could be accounted for by a faster CPU? I don't suppose you have a link to that workunit?

That particular CPU I was referring to is the wingman of my work unit in this post.
But scrolling through your 3x list I've seen an i7 take over 200,000 seconds and an i5 wingman trounce it by doing the same WU in 3x short time.
My slower clocked XEON CPU taking much less time than a faster clocked XEON CPU (2.53GHz v's 3.4 GHz).
So the processor may not solely be the reason for the far different times being seen.

Conan
____________

Profile Conan
Avatar
Send message
Joined: 13 Apr 12
Posts: 151
Combined Credit: 47,672,899
DNA@Home: 399,792
SubsetSum@Home: 1,448,876
Wildlife@Home: 45,824,231
Wildlife@Home Watched: 70,910s
Wildlife@Home Events: 0
Climate Tweets: 412
Images Observed: 0

          
Message 6909 - Posted: 15 Apr 2017, 10:35:39 UTC - in response to Message 6907.

Had this one and both myself and my wingman had large run times but the credit seems a bit low for the run time this wu

Mine ran over 260,000 seconds and my wingman ran over 332,000 seconds for 1,911 credits.

Conan


Thoughts on what might be more appropriate for that kind of runtime?


Just going on what some of my other much shorter work units are getting then the awarded credit would have to be over 7,000 points (I have seen a number of work units around the 70,000 second mark get 5,000+ points, I have a few myself).
So 7,000 would be a lower amount for these much larger work units and then doing what I suggest below.

The multiplier on some of the short ones if it was applied against the larger ones would see you handing out 20,000 to 30,000 points a work unit which is getting a bit high don't you think?

I suppose you could just set a maximum limit you are willing to pay, say 10,000 points and no matter how long the work units then run they will get 10,000 points, which is still a good return.
You just would have to work out when that limit cuts in with a scaling of awarded points up to the limit then a set amount.

I have 4 large work units with very ordinary credit, ranging from 10 to 28 cr/h

1009561 98.82h / 2825.5cr = 28.59 cr/h
1008772 114.55h / 3218.04cr = 28.09 cr/h
1001360 92.46h / 1911.57cr = 20.67 cr/h
1000879 145.56h / 1496.97cr = 10.28 cr/h

When you get some of the short ones getting 300cr/h then the long ones should get a better deal.

Just some thoughts.
Thanks Travis

Conan


I just did a quick calculation on my currently long running work unit (estimated over 11 days) and 10,000 points whilst reasonable would still be low, especially when compared to shorter work units, so perhaps a ceiling that is higher than this? Maybe up to 20,000 but you would still have to really look into this.

I would not like a lot of people dumping these long runners when they think they are not getting a good reward for them.

Thanks again

Conan
____________

Travis Desell
Volunteer moderator
Project administrator
Project developer
Project scientist
Send message
Joined: 16 Jan 12
Posts: 1813
Combined Credit: 23,514,257
DNA@Home: 293,563
SubsetSum@Home: 349,212
Wildlife@Home: 22,871,482
Wildlife@Home Watched: 212,926s
Wildlife@Home Events: 51
Climate Tweets: 21
Images Observed: 774

              
Message 6910 - Posted: 15 Apr 2017, 19:28:19 UTC - in response to Message 6907.

Had this one and both myself and my wingman had large run times but the credit seems a bit low for the run time this wu

Mine ran over 260,000 seconds and my wingman ran over 332,000 seconds for 1,911 credits.

Conan


Thoughts on what might be more appropriate for that kind of runtime?


Just going on what some of my other much shorter work units are getting then the awarded credit would have to be over 7,000 points (I have seen a number of work units around the 70,000 second mark get 5,000+ points, I have a few myself).
So 7,000 would be a lower amount for these much larger work units and then doing what I suggest below.

The multiplier on some of the short ones if it was applied against the larger ones would see you handing out 20,000 to 30,000 points a work unit which is getting a bit high don't you think?

I suppose you could just set a maximum limit you are willing to pay, say 10,000 points and no matter how long the work units then run they will get 10,000 points, which is still a good return.
You just would have to work out when that limit cuts in with a scaling of awarded points up to the limit then a set amount.

I have 4 large work units with very ordinary credit, ranging from 10 to 28 cr/h

1009561 98.82h / 2825.5cr = 28.59 cr/h
1008772 114.55h / 3218.04cr = 28.09 cr/h
1001360 92.46h / 1911.57cr = 20.67 cr/h
1000879 145.56h / 1496.97cr = 10.28 cr/h

When you get some of the short ones getting 300cr/h then the long ones should get a better deal.

Just some thoughts.
Thanks Travis

Conan


Hmm, I think there may be some things going on with my credit calculation that I'm not taking into account if short WUs are getting 300cr/hr and long ones are getting 30cr/hr. I'm wondering if cache sizes or some other hardware aspects are causing things.

I'm working on getting some code in there so I can plot runtime vs workunit size to get a better overall picture of what's going on.

Profile Conan
Avatar
Send message
Joined: 13 Apr 12
Posts: 151
Combined Credit: 47,672,899
DNA@Home: 399,792
SubsetSum@Home: 1,448,876
Wildlife@Home: 45,824,231
Wildlife@Home Watched: 70,910s
Wildlife@Home Events: 0
Climate Tweets: 412
Images Observed: 0

          
Message 6912 - Posted: 16 Apr 2017, 23:58:18 UTC - in response to Message 6910.
Last modified: 16 Apr 2017, 23:59:25 UTC

Had this one and both myself and my wingman had large run times but the credit seems a bit low for the run time this wu

Mine ran over 260,000 seconds and my wingman ran over 332,000 seconds for 1,911 credits.

Conan


Thoughts on what might be more appropriate for that kind of runtime?


Just going on what some of my other much shorter work units are getting then the awarded credit would have to be over 7,000 points (I have seen a number of work units around the 70,000 second mark get 5,000+ points, I have a few myself).
So 7,000 would be a lower amount for these much larger work units and then doing what I suggest below.

The multiplier on some of the short ones if it was applied against the larger ones would see you handing out 20,000 to 30,000 points a work unit which is getting a bit high don't you think?

I suppose you could just set a maximum limit you are willing to pay, say 10,000 points and no matter how long the work units then run they will get 10,000 points, which is still a good return.
You just would have to work out when that limit cuts in with a scaling of awarded points up to the limit then a set amount.

I have 4 large work units with very ordinary credit, ranging from 10 to 28 cr/h

1009561 98.82h / 2825.5cr = 28.59 cr/h
1008772 114.55h / 3218.04cr = 28.09 cr/h
1001360 92.46h / 1911.57cr = 20.67 cr/h
1000879 145.56h / 1496.97cr = 10.28 cr/h

When you get some of the short ones getting 300cr/h then the long ones should get a better deal.

Just some thoughts.
Thanks Travis

Conan


Hmm, I think there may be some things going on with my credit calculation that I'm not taking into account if short WUs are getting 300cr/hr and long ones are getting 30cr/hr. I'm wondering if cache sizes or some other hardware aspects are causing things.

I'm working on getting some code in there so I can plot runtime vs workunit size to get a better overall picture of what's going on.


Just to add This WU ran for 373236 seconds (103 hours) on my XEON at 2.53 GHz, yet my wingman ran for 92,103 seconds on a XEON at 2.60 GHz. Awarded 3345.19 credits, I ended up with 32 cr/h and the wingman very much higher.
So Processor does not seem to be the reason run times vary so much.

Conan
____________

Travis Desell
Volunteer moderator
Project administrator
Project developer
Project scientist
Send message
Joined: 16 Jan 12
Posts: 1813
Combined Credit: 23,514,257
DNA@Home: 293,563
SubsetSum@Home: 349,212
Wildlife@Home: 22,871,482
Wildlife@Home Watched: 212,926s
Wildlife@Home Events: 51
Climate Tweets: 21
Images Observed: 774

              
Message 6913 - Posted: 17 Apr 2017, 3:46:34 UTC - in response to Message 6912.

Had this one and both myself and my wingman had large run times but the credit seems a bit low for the run time this wu

Mine ran over 260,000 seconds and my wingman ran over 332,000 seconds for 1,911 credits.

Conan


Thoughts on what might be more appropriate for that kind of runtime?


Just going on what some of my other much shorter work units are getting then the awarded credit would have to be over 7,000 points (I have seen a number of work units around the 70,000 second mark get 5,000+ points, I have a few myself).
So 7,000 would be a lower amount for these much larger work units and then doing what I suggest below.

The multiplier on some of the short ones if it was applied against the larger ones would see you handing out 20,000 to 30,000 points a work unit which is getting a bit high don't you think?

I suppose you could just set a maximum limit you are willing to pay, say 10,000 points and no matter how long the work units then run they will get 10,000 points, which is still a good return.
You just would have to work out when that limit cuts in with a scaling of awarded points up to the limit then a set amount.

I have 4 large work units with very ordinary credit, ranging from 10 to 28 cr/h

1009561 98.82h / 2825.5cr = 28.59 cr/h
1008772 114.55h / 3218.04cr = 28.09 cr/h
1001360 92.46h / 1911.57cr = 20.67 cr/h
1000879 145.56h / 1496.97cr = 10.28 cr/h

When you get some of the short ones getting 300cr/h then the long ones should get a better deal.

Just some thoughts.
Thanks Travis

Conan


Hmm, I think there may be some things going on with my credit calculation that I'm not taking into account if short WUs are getting 300cr/hr and long ones are getting 30cr/hr. I'm wondering if cache sizes or some other hardware aspects are causing things.

I'm working on getting some code in there so I can plot runtime vs workunit size to get a better overall picture of what's going on.


Just to add This WU ran for 373236 seconds (103 hours) on my XEON at 2.53 GHz, yet my wingman ran for 92,103 seconds on a XEON at 2.60 GHz. Awarded 3345.19 credits, I ended up with 32 cr/h and the wingman very much higher.
So Processor does not seem to be the reason run times vary so much.

Conan



Guh, this is really weird. I'm gathering data so I can try and get an idea what on earth might be causing this.

Profile Saenger
Avatar
Send message
Joined: 28 Apr 13
Posts: 23
Combined Credit: 2,269,715
DNA@Home: 0
SubsetSum@Home: 6,825
Wildlife@Home: 2,262,890
Wildlife@Home Watched: 64,483s
Wildlife@Home Events: 0
Climate Tweets: 0
Images Observed: 0

    
Message 7023 - Posted: 3 May 2017, 14:40:35 UTC

Regardless of the credits, I've just got one monster on my machine, seems to be really one, as my wingman is not ready as well, but now there are two more victims, as the deadline has just passed with both of us.
This is the WU
This are the WUs form my machine

It's an EXACT MNIST Convolutional Neural Network Trainer v0.20, and the data from my WU are currently:

CPU-Time last Checkpoint: 83:45:05 current CPU-Time: 83:53:43 estimated remainder: 11:36:58 (which I do not believe ;) Process: 85,657%

Is there anything I could do but to wait and see?
____________
Grüße vom Sänger

Previous · 1 · 2 · 3 · 4 · Next
Post to thread

Message boards : Number Crunching : Monster MNIST units